Hiding from Facebook: An Encryption Protocol resistant to Correlation Attacks

TL;DR

The paper addresses privacy leakage in social networks where public posts can reveal hidden identifiers through correlation attacks. It introduces a per-instance encryption pipeline that randomizes hidden content to produce decorrelated representations, measured with Canonical Correlation Analysis, and secures them with symmetric encryption whose key is distributed to a trusted group via a multi-party Diffie-Hellman protocol. The authors prove security under the Computational Diffie-Hellman and related hypotheses and discuss practical limitations, including active man-in-the-middle threats that require authentication. Overall, it offers a principled, privacy-preserving framework for tagging in social networks, balancing decorrelation strength with key management and deployment constraints.

Abstract

In many social networks, one publishes information that one wants to reveal (e.g., the photograph of some friends) together with information that may lead to privacy breaches (e.g., the name of these people). One might want to hide this sensitive information by encrypting it and sharing the decryption key only with trusted people, but this might not be enough. If the cipher associated to a face is always the same, correlation between the output of a face recognition system and the cipher can give useful clues and help train recognizers to identify untagged instances of the face. We refer to these as "correlation attacks". In this paper we present a coding system that attempts to counter correlation attacks by associating to each instance of a face a different encryption of the same tag in such a way that the correlation between different instances is minimal. In addition, we present a key distribution code that allows only the owner of the images to encode the tags, but allows a group of trusted friends to decode them.

Figures (4)

• Figure 1: The general encryption schema. The manifest content $m$ is passed as is to the public domain. The hidden content $h$ is first scrambled with the aid of a random number $r$ so that different copies of the same manifest content $m$ will be associated to different, and uncorrelated, contents $u$. The content $u$ is then encrypted to obtain the code $c(u)$, which is released in the public domain.
• Figure 2: Encoding the numbers $0,\ldots,4$ with three possibility of code each. The base for the code is the number $p=10$, after which there are four slots, one for each number. (At least) four slots are present after the numbers $p_5=15$, $p_6=21$, and $p_7=28$. For example, the number 2 can be encoded as 12, 17, or 23.
• Figure 3: The correlation between $h$ and $u$ with $b$ and $B$ bits, respectively, for the quadratic randomization scheme (insertion of random bits in fixed position results in a correlation of $1$ in all cases). The curves are parametrized by the number of bits of $h$ and plotted as a function of $B/b$, where $B\ge{2b}$ is the number of bits of $u$.
• Figure 4: Correlation of the encrypted hidden content with the original hidden content when the randomization has been obtained with insertion of random bits in fixed positions (line with $\lozenge$) and with the method presented here (line with $\blacklozenge$).

Theorems & Definitions (2)

• Theorem 4.1
• proof